ResearchAI model Astra reportedly cracked an 85-year-old Enigma message in two daysResearchers at the independent lab Crypto Cellar say OpenAI's new model Astra (referred to as GTP-Astra) decrypted a previously unsolved Enigma ciphertext in two days.2 min read
ResearchAI reshapes science and medicine while transforming jobs, studies showDebate about AI often centres on job losses, but recent studies document substantial benefits in medicine and scientific research alongside complex labour-market effects.4 min read
ResearchWhen Ads Compete with Advice: Study Shows Many LLMs Favor Sponsored OptionsA new academic study accepted to COLM 2026 tests how large language models behave when a platform’s commercial incentives conflict with a user’s interest.4 min read
ResearchHow Digitalization Is Reshaping Our Relationship to Physical SpaceDigitalization has blurred the boundaries between online and offline spaces, altering everyday behaviours, urban form and social relations.4 min read
ResearchClaude AI performed nine-loop computation in the planar N=4 super-Yang–Mills modelAnthropic’s Claude artificial intelligence solved the nine-loop scattering amplitude calculation in the planar N=4 super-Yang–Mills model after running for several days in the Claude Science environment in response to a challenge from a month ago.1 min read
ResearchThree Anthropic AI Scenarios: Big GDP Gains, Significant Social Costs PossibleAnthropic’s economics team published three modeled scenarios showing how AI could affect the U.S.4 min read
ResearchComparative test: Astra topped the models' auditing rankingsIn a comparative test, several new large language models each produced an audit, and a panel of judges ranked the works.1 min read
ResearchUnified multi-agent system for coherent, multi-minute video generationResearchers present an orchestration-layer, multi-agent framework that plans and enforces visual continuity across multi-shot narratives, built atop foundation models Gemini and Veo.5 min read
ResearchOpus 5.5 highlighted the need to optimize shell startup timeIn ShellPerfBench, the Opus 5.5 model found shell startup time optimizations that other models missed; this can affect users' .zshrc and other dotfiles and quietly slow down opening the terminal, so…1 min read
ResearchMentalHealthBench extends its mental-health AI assessment to everyday supportAs a new benchmark, the developers of MentalHealthBench aim to cover the full mental-health spectrum of people turning to artificial intelligence — from everyday support to severe crises.1 min read
ResearchClaude discovered a previously unknown enzyme system in bacteriophage DNAThe enzyme system discovered by Claude is located in bacteriophage DNA, adjacent to a long, repetitive DNA sequence resembling CRISPR.1 min read
ResearchStudy: Offloading GPU inference to edge or cloud improves mobile robot performance and battery lifeA systematic evaluation of mobile manipulation workloads shows that performing AI inference exclusively on onboard GPUs can degrade task success, limit model size, and shorten battery life.4 min read
